Many thanks!

We have received your enquiry and will be in touch as soon as possible.

SwissGlobal Newsletter

Lost in translation? Not with SwissGlobal. Stay up to date with language industry news, tips, interviews and more – subscribe to our monthly newsletter.

Machine translation research project: SwissGlobal receives funding from Innosuisse

SwissGlobal, in collaboration with the Università della Svizzera italiana (USI) and the University of Applied Sciences and Arts Northwestern Switzerland (FHNW), has received funding from Innosuisse for the development of a toolchain for neural machine translation (NMT). The toolchain should result in significant qualitative improvements and increase efficiency in the implementation and operation of translation engines. With this project, SwissGlobal is proving that high quality standards, innovation and efficiency are not contradictions when it comes to language services.

As the internet and globalisation continue to drive the translation market forward, the market for machine translation is experiencing particularly dynamic development, with annual growth of 19% expected for the 2020-2024 period, according to the online magazine Business Insider. Machine translation (MT) is making it possible to translate large volumes of text that were previously only available in a single language due to money or time constraints. Whereas translations produced by the first-generation MT programmes were often unintentionally comical, those completed by neural machine translators (NMT) are already delivering surprisingly good results. Google Translate and DeepL are just two of the most prominent examples of NMT translation solutions. You can read our blog post “Neuronal machine translation: learning by doing” to find out exactly how NMT works.

Neural machine translation not yet feasible for Swiss SMEs

However, these innovations still provide too few benefits when it comes to Swiss small and medium-sized enterprises (SMEs). Open-source machine translators such as DeepL and Google Translate are free or offer low-cost professional subscriptions. Although it may sound good, free services do not provide any data security, and subscriptions usually do not offer specialised terminology.

This leaves many companies with the sole option of purchasing their own translation engine and training it to meet their specific terminological needs and language combinations. This approach guarantees better translation quality as well as security for proprietary content, but it is also extremely time-consuming and very expensive. Specialised multilingual datasets for training a translation engine, such as offered by the data network TAUS, cost up to 60,000 euros per year! This is where SwissGlobal’s innovation project bridges a crucial gap, both in terms of data security and costs as well as usability for SMEs.

Innovation project headed by SwissGlobal

SwissGlobal is collaborating with the University of Applied Sciences and Arts Northwestern Switzerland (FHNW) and the Università Svizzera italiana (USI) on a new type of toolchain that will feature many advantages and innovations compared to both the free and fee-based machine translation technology that is currently available. The proprietary toolchain combines intelligent technologies to create a revolutionary, almost fully automated application. Its innovative strength lies in the efficient collection of high-quality data and the development of innovative algorithms for training the translation engine. The aim is to provide clients with a flexible, high-quality and innovative solution that fulfils their requirements for specific sectors and languages.

Federal funding for “turbo data”

Training accounts for the greatest expenditure when setting up a customised translation engine. This is because the algorithms of a translation machine need large amounts of data (known as “translation units”) during training to produce an acceptable translation result. Working with Professor Rolf Krause’s team at USI, various innovative approaches are now being developed to significantly accelerate the training process. The research project leader, Professor Manfred Vogel, heads the Information Processing department at the Institute for Data Science at FHNW, while Professor Rolf Krause is a computer science professor at USI and conducts research in the field of machine learning, among other things. Both institutes are leaders in their respective fields.

The toolchain created by SwissGlobal, FHNW and USI has also impressed Innosuisse, the Swiss Innovation Agency. Innosuisse is a federal entity under public law that promotes science-based innovation in the interest of the economy and society in Switzerland. It approved funding for the development project in early August.

Meeting specific client needs with customised innovation

The goal is to have a deployable, proprietary translation engine ready by spring 2023 that is scalable and can be quickly trained for client-specific needs and projects. The quality of the translations should be higher than what is achieved with current standard machine translation systems. Another priority is ensuring a high level of data security. The engine will be hosted either on SwissGlobal’s servers in Switzerland or, if necessary, directly on the client’s premises, making it possible to meet the highest requirements in terms of data security. After all, technology is only beneficial when the interface between human skill and machine capability functions smoothly. SwissGlobal uses technology to ensure faster and more efficient processes, the secure transfer of data, and high-quality, consistent translations. At SwissGlobal, all of our systems and processes are certified to ISO 9001 (quality management systems), ISO 17100 (translation services) and, most recently, to ISO 18587 (post-editing) as well. Do you have any questions? Please contact us, we would be happy to help.